Information Technology Reference
In-Depth Information
Fig. 5.32 Connection matrix for the chemical neural network (a) and recognition of the numerical
sequence (b)
random sets of numbers). These reactors were connected by mass transfer. The
connection matrix for this system is shown in Fig. 5.32 .
Assume that some initial distribution of concentrations is introduced into the
system that is different from the state of the reactors being stored when their
connections are disabled. Then, after the connections are activated, there are two
possibilities:
￿ The network of reactors will have a uniform (homogeneous) distribution of
states, if the input image is significantly different from the one being stored.
￿ One of the stored images appears in the network if the input image is close to one
of the images being stored.
The efficiency of such a chemical network was tested with a set of three different
images slightly different from the ones stored in the network structures. It was thus
shown that
the efficiency of recognition of the structures is quite high (see
Fig. 5.32 ).
The investigation of the possibilities of pattern recognition by chemical net-
works, made by the authors [139], was the first experimental study in this area. It
should nevertheless be noted that, despite the undoubted importance of this work,
its technical solution was too complex. The equipment used for pattern recognition
was cumbersome and inconvenient to operate effectively.
Other solutions for creating chemical multilevel systems for solving pattern
recognition problems were proposed. A promising example is the idea of electro-
chemical connection between subsystems. In general, the possibilities of this
approach are far from exhausted, and we can expect in the near future the emer-
gence of new variants of chemical recognition systems.
Search WWH ::




Custom Search